Salesforce AI token pricing has moved from a technical curiosity to a balance-sheet variable that investors in software and infrastructure stocks can no longer ignore. As artificial intelligence workloads shift from simple queries to multi-step agentic tasks, the economics of who pays for computation, and how, are reshaping gross margins across the sector.
From Seat Licences to Consumption: What Salesforce AI Token Pricing Actually Means
The traditional software model sold access by the seat. A company bought 500 licences and the vendor’s revenue was largely predictable. Agentic AI breaks that contract: an AI agent completing a complex workflow may consume thousands of tokens, running costs that neither vendor nor customer anticipated at contract signing.
Salesforce has moved deliberately to address this. By May 2025, the company had introduced Flex Credits, priced at $500 for 100,000 credits, with each standard AI action costing 20 credits, equivalent to roughly $0.10 per action. That is a consumption-based structure: customers pay for what the agent actually does, rather than for seats that may sit idle. In December 2025, chief executive Marc Benioff told analysts that seat-based pricing was nonetheless becoming the norm for AI agents, pointing to Salesforce’s new Agentic Enterprise License Agreement as evidence that customers were demanding cost predictability alongside flexibility.
Those two positions are not contradictory. The Flex Credits model gives smaller or variable workloads a pay-as-you-go option, while the enterprise licence agreement gives large organisations a budgetable ceiling. Together, they represent a hybrid billing architecture designed to preserve Salesforce’s gross margin while giving procurement teams something they can model.
The Margin Risk for Software Companies That Have Not Adapted
Not every software vendor has moved as quickly. For companies still billing primarily by seat while their underlying AI costs scale with usage, the arithmetic is uncomfortable. Each agentic task carries a compute cost, sometimes called the “token tax,” that flows directly into cost of goods sold. If revenue is fixed by a seat licence but costs rise with AI adoption, gross margins compress. Over a five-to-ten-year horizon, that compression is likely to be a persistent feature of any software business that has not restructured its pricing model.
Adobe has similarly explored consumption or hybrid models for its AI features, according to the original analysis. Infrastructure businesses, including Nvidia, Micron and Broadcom, sit on the other side of this equation: rising token demand is a tailwind for hardware and memory, regardless of how the application layer prices its product.
For a UK investor managing a self-invested personal pension (SIPP) or a balanced portfolio, the distinction matters. Infrastructure names offer exposure to AI demand that is relatively agnostic to software pricing structures, though they carry their own cyclicality and valuation risk. Software names require closer scrutiny of gross margin trends in quarterly filings, not just revenue growth, before any position is sized.
What Portfolio Managers Should Watch
The signal to monitor is gross margin direction, reported each quarter. A software company growing its AI revenue while maintaining or expanding gross margins has likely built the pricing architecture to absorb rising compute costs. One whose gross margins are drifting lower as AI adoption rises is effectively subsidising its customers’ AI usage, a position that becomes harder to sustain as agentic workloads deepen.
Salesforce‘s Flex Credits mechanism is a live experiment in whether granular, action-level pricing can preserve margins at scale. The pricing data published by Next Waves Insight puts a concrete number on that experiment: $0.10 per standard AI action. Investors who track whether that unit cost holds, rises, or falls as volumes grow will have a more grounded view of Salesforce’s margin trajectory than those watching only headline revenue. For anyone in the accumulation phase with a ten-year-plus horizon, that margin trajectory is likely to matter more than the current quarter’s bookings figure.
